An Efficiently Computable Metric for Multiple Extended Target Tracking

Abstract

In multiple extended target tracking, the aim is to estimate the positions, shapes and trajectories of extended targets, as well as the number of trajectories. The commonly used metrics are not sufficient to comprehensively evaluate the performance of the multiple extended target tracking. The major challenge of the metric is that it takes into account the similarity of extended targets in the same trajectory over a period of time and the dissimilarity between different target trajectories. In this paper, a new mathematically metric is proposed to jointly measure position, shape and label estimates, which can capture errors such as track separation and switching. The position and shape errors are firstly used to measure the same extended target in continuous time, and then the optimal sub-pattern assignment distance is used to measure the errors between the overall target trajectories. Simulation results verify that the proposed metric can capture differences between two sets of extended target trajectories.